| In recent years, with the rapid development of power electronics,power systems have emerged many serious power quality problems, suchas harmonic pollution, reactive power, voltage fluctuations andimbalances. Active Power Filter (APF) is recognized as one effective wayto control the harmonic and reactive power pollution and it has become arelatively new research focus. However, the application of APF in ourcountry is far from mature compared to the passive filter, there are stillmany issues need to be further studied and improved.APF can eliminate harmonic current by generating compensatingcurrent which has the same amplitude and opposite phase with sourceharmonious wave. In order to track the reference current quickly andaccurately, design of controller plays an important role. Considering thatthe reference signal is periodic and the traditional repetitive controller hasa cycle delay, the Neural Network PI Repetitive Controller (NNPIRC) isproposed in this thesis. The Neural Network is adopted to improve response speed by tuning the PI parameters adaptively, making up delaydefect of the repetitive controller, while the repetitive controller is used toimprove tracking accuracy. However, because the NNPIRC must betrained to optimize the parameters in each sample time, it cost lots ofcalculation and time. Considering that Recursive Integral PI controller isequivalent to N PI controllers working in parallel, and it just need to betuned in one period, so it can greatly improve the rapidity and accuracy.Therefore the Neural Network Recursive Integral PI Repetitive Controller(NNRIPIRC) is designed, it uses neural network to tune recursive integralPI parameters on-line, and takes advantage of the repetitive controller toimprove the tracking accuracy. As a result, the initial tracking speed andaccuracy has been improved significantly by using the NNRIPIRC thanthe controller above. However, in practice, the three-phase power load isalways asymmetry which causes negative sequence current pouring intothe power system, and those controllers proposed above can't achievegood results in this situation. In order to eliminate the negative sequencecurrent, the Neural Network Recursive Integral PI Repetitive Controllerwith Decoupling of positive and negative sequence (NNRIPIRC-D) ispresented, it can control positive and negative sequence currentindependently and inhibit twice frequency harmonic of DC side. So it cansolve the problem when the power grid is asymmetric, thereby theNNRIPIRC-D improves overall system performance. Finally, simulation results demonstrate the effectiveness of the method. |
